---

一、问题

在使用Java向ES库中插入数据时，第一次成功插入，第二次出现以下错误：
java.lang.IllegalStateException: Request cannot be executed; I/O reactor status: STOPPED at

问题原因

这里显示是连接中断，第一次遇到这个问题，比较疑惑为什么es的客户端会中断，理论上es client 是长连接，不停的有数据写入，连接一致存在，除非有服务端异常。在elasticsearch服务端查看日志，没有任何异常信息。

网上搜索错误信息原来是 Apache HTTPComponents 异步客户端问题。es官网有个issues详细的记录的这个问题的原因和修复建议。

官方的意见是 Apache HTTPComponents 异步客户端 使用了一个内部的I/O reactor 分发IO event。在某些情况下，IO reactor会记录程序调用栈中的异常或者Java NOI库中的异常，如果这些异常不被处理，I/O reactor会直接关闭，es client不可用，此时只能重启服务。es client中试图增加一个默认的 I/O reactor 异常处理逻辑但是在做了一些尝试后发现捕获I/O reactor后会导致SSL中断。而HTTPComponents 在版本5中已经修复了这个问题，最终官网给的建议是等待版本升级。

二、解决思路

解决问题的过程中参考了以下文档：
https://www.cnblogs.com/yangchongxing/p/15440197.html
https://github.com/elastic/elasticsearch/issues/42133
https://zhuanlan.zhihu.com/p/384269417
https://cloud.tencent.com/developer/article/1806886

主要获得解决方法的是以下：
https://github.com/elastic/elasticsearch/issues/39946
主要引用以下：

大概意思是说:
在每个线程需要时创建一个新的客户端，并在方法结束时关闭。这就解决了问题。
结合GPT获取解决方案：

import java.util.concurrent.Executors;
import java.util.concurrent.ExecutorService;public class EsClientManager {private final ExecutorService executorService = Executors.newFixedThreadPool(10);private final ThreadLocal<RestHighLevelClient> CLIENT_THREAD_LOCAL = ThreadLocal.withInitial(() -> {RestHighLevelClient client = null;try {client = RestHighLevelClient.builder(new HttpHost("localhost", 9200, "http")).build();return client;} catch (IOException e) {e.printStackTrace();return null;}});public RestHighLevelClient getClient() {return CLIENT_THREAD_LOCAL.get();}public void closeClient(RestHighLevelClient client) {CLIENT_THREAD_LOCAL.remove();if (client != null) {executorService.execute(() -> {try {client.close();} catch (IOException e) {e.printStackTrace();}});}}
}

在使用时，同样可以这样：

public class MyService {public void someMethod() {RestHighLevelClient client = EsClientManager.getClient();try {// 执行操作} finally {EsClientManager.closeClient(client);}}
}

这样，每个线程都会从EsClientManager获取一个客户端，并在方法结束时自动关闭。

线程池不允许使用 Executors 去创建，而是通过 ThreadPoolExecutor 的方式，进一步，优化：

import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.ThreadPoolExecutor;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.LinkedBlockingQueue;public class EsClientManager {private final ExecutorService executorService;private final ThreadLocal<RestHighLevelClient> CLIENT_THREAD_LOCAL = ThreadLocal.withInitial(() -> {RestHighLevelClient client = null;try {client = RestHighLevelClient.builder(new HttpHost("localhost", 9200, "http")).build();return client;} catch (IOException e) {e.printStackTrace();return null;}});public EsClientManager() {// 配置 ThreadPoolExecutorint corePoolSize = 10; // 核心线程数int maximumPoolSize = 10; // 最大线程数long keepAliveTime = 0L; // 空闲线程等待新任务的最长时间TimeUnit unit = TimeUnit.MILLISECONDS; // keepAliveTime的时间单位int queueCapacity = 100; // 工作队列的容量ThreadPoolExecutor executor = new ThreadPoolExecutor(corePoolSize,maximumPoolSize,keepAliveTime,unit,new LinkedBlockingQueue<>(queueCapacity), // 工作队列runnable -> {Thread thread = new Thread(runnable);thread.setDaemon(false); // 设置线程是否为守护线程，false表示非守护线程return thread;});this.executorService = Executors.unconfigurableExecutorService(executor);}public RestHighLevelClient getClient() {return CLIENT_THREAD_LOCAL.get();}public void closeClient(RestHighLevelClient client) {CLIENT_THREAD_LOCAL.remove();if (client != null) {executorService.execute(() -> {try {client.close();} catch (IOException e) {e.printStackTrace();}});}}// 添加方法以允许关闭executorServicepublic void shutdown() {executorService.shutdown();}public boolean isShutdown() {return executorService.isShutdown();}public boolean isTerminated() {return executorService.isTerminated();}
}